What it does
Aide Decision Pause is a privacy-conscious companion for moments when emotion and urgency can push someone toward a consequential choice. It gently gathers context one question at a time, then produces an inspectable Decision Pause: known facts versus assumptions, plausible future paths, reversibility, uncertainty, and a low-regret next action.
How it works
The Build Week flow uses GPT-5.6 through the OpenAI Responses API in two stages:
- Clarification uses low reasoning effort and asks one gentle question per turn.
- Final guidance uses medium reasoning by default.
- High reasoning is available only after an explicit user choice.
- Requests use
store: false.
Deterministic safety and profile processing runs before generation. The final artifact can draw only from active, rights-cleared practical-wisdom records, with provenance carried through retrieval. Consented memory, bounded context, structured summaries, telemetry, and evaluation gates support privacy, cost control, and reliability.
What was built during Build Week
The repository already had a TypeScript monorepo, Fastify API, PWA, safety-classifier boundary, consent-aware memory, RAG abstractions, ingestion and evaluation harnesses, persistence adapters, workers, and infrastructure overlays before Build Week.
The Build Week extension adds the native GPT-5.6 Responses API provider, staged clarification and guidance, the Decision Pause structure and prompts, a license-safe core-wisdom corpus, targeted tone/routing/reasoning/grounding/safety/rights evals, and updated transparency, privacy, deployment, demo, and release documentation.
Why it matters
Aide focuses on the gap between emotional impulse and consequential action. It is not a generic chatbot: its core product primitive is a deliberate pause that keeps the person in control while making evidence, uncertainty, alternatives, and reversibility visible.
Built with Codex
Codex was used to implement and test the Build Week extension across the monorepo, including the OpenAI provider, staged orchestration, prompts, user experience, provenance controls, and targeted release checks. The final submission will include the primary /feedback session ID and dated repository evidence after those items are verified.
Built With
- codex
- docker
- fastify
- gpt-5.6
- openai-responses-api
- pgvector
- postgresql
- qdrant
- react
- typescript
- vite